Routing and Transport in Wireless Sensor Networks Ibrahim Matta - - PDF document

▶
routing and transport in wireless sensor networks
SMART_READER_LITE
LIVE PREVIEW

Routing and Transport in Wireless Sensor Networks Ibrahim Matta - - PDF document

Dec 16, 2022 390 likes •635 views

10/21/2003 Routing and Transport in Wireless Sensor Networks Ibrahim Matta (matta@bu.edu) Niky Riga (inki@bu.edu) Georgios Smaragdakis (gsmaragd@bu.edu) Wei Li (wli@bu.edu) Vijay Erramilli (evijay@bu.edu) References Adaptive Protocols

slide-1
SLIDE 1

10/21/2003

Routing and Transport in Wireless Sensor Networks

Ibrahim Matta (matta@bu.edu) Niky Riga (inki@bu.edu) Georgios Smaragdakis (gsmaragd@bu.edu) Wei Li (wli@bu.edu) Vijay Erramilli (evijay@bu.edu)

10/07/2003 Ibrahim Matta

References

  • Adaptive Protocols for Information Dissemination in Wireless Sensor Networks

Wendi Rabiner Heinzelman, J. Kulik, and H. Balakrishnan Proceedings of the Fifth Annual ACM/IEEE International Conference on Mobile Computing and Networking (MobiCom 1999), Seattle, Washington, August 15-20, 1999, pp. 174-185.

  • Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks

Chalermek Intanagonwiwat, Ramesh Govindanand Deborah Estrin Proceedings of the Sixth Annual International Conference on Mobile Computing and Networks (MobiCOM 2000), August 2000, Boston, Massachusetts.

  • Rumor Routing Algorithm For Sensor Networks

David Braginsky and Deborah Estrin First Workshop on Sensor Networks and Applications (WSNA), September 28, 2002, Atlanta, GA.

  • Highly Resilient, Energy Efficient Multipath Routing in Wireless Sensor Networks

Deepak Ganesan, Ramesh Govindan, Scott Shenker and Deborah Estrin Mobile Computing and Communications Review (MC2R), Vol 1., No. 2. 2002.

  • GRAdientBroadcast: A Robust Data Delivery Protocol for Large Scale Sensor Networks

Fan Ye, Gary Zhong, SongwuLu, LixiaZhang ACM WINET (Wireless Networks)

  • Energy-efficient Communication Protocol for Wireless Microsensor Networks

Wendi Heinzelman, Anantha Chandrakasan, Hari Balakrishnan Proceedings of the Hawaii International Conference on Systems Science, January 2000, Maui, HI.

  • A Two-tier Data Dissemination Model for Large-scale Wireless Sensor Networks

Fan Ye, Haiyun Luo, Jerry Cheng, Songwu Lu, LixiaZhang Proceedings of the Eighth Annual ACM/IEEE International Conference on Mobile Computing and Networking (MobiCOM 2002), September 2002, Atlanta, GA.

  • PSFQ: A Reliable Transport Protocol For Wireless Sensor Networks

Chieh-YihWan, Andrew Campbell, Lakshman Krishnamurthy First Workshop on Sensor Networks and Applications (WSNA),September 28, 2002, Atlanta, GA.

slide-2
SLIDE 2

10/21/2003

10/07/2003 Ibrahim Matta

Model

  • Sensor nodes perform sensing tasks and report back

data to user (via the “sink”)

  • Sensor nodes are resource-constrained (limited battery

power, processing power, memory, etc.)

  • High transmission error rate and low bandwidth when

nodes communicate over wireless

10/07/2003 Ibrahim Matta

Model (cont’d)

  • Data flowing from sources (sensors) to “sink” is usually

loss-tolerant

– E.g., sensing temperature, light, acoustic, etc.

  • Data flowing from “sink” to sensors is usually loss-sensitive

– E.g., sensor management: re-tasking or re-programming sensors

slide-3
SLIDE 3

10/21/2003

10/07/2003 Ibrahim Matta

Application-specific Protocols

  • Data aggregation opportunities

– Remove duplicate or redundant data – “Beamforming” or fusion

  • Routing and transport intertwined

– Data centric

  • Want a long-lived, robust, low-latency

network …

– that scales to large number of sensors, sinks, and high mobility

10/07/2003 Ibrahim Matta

Protocol Design Goals

  • Low Energy

– Minimize communication

Aggregate data in network

– Low Node Duty Cycle

Minimize individual node responsibility Traffic spreading / Load balancing Shut down nodes when possible

  • Robust

– Adapt to unpredictable environment without intervention

  • Scalable

– Rely on localized algorithms –no centralized control

  • Low Latency

– Must meet application latency and accuracy requirements

  • Small Footprint

– Must run on hardware with severe memory and computational power constraints

slide-4
SLIDE 4

10/21/2003

10/07/2003 Ibrahim Matta

Example Network Models

Broadcast Unicast Multicast Static

Interest Propagation

Target Detection Continuous Query Broadcast Multicast Unicast

Data Dissemination

Mobile Mobile Stationary Stationary

Event Users (Sinks) Sensors (Sources)

10/07/2003 Ibrahim Matta

Protocols

  • Flooding
  • Gradient
  • Clustering
  • Reliable
  • Geographic
slide-5
SLIDE 5

10/21/2003

10/07/2003 Ibrahim Matta

Flooding Based Approaches

  • Flooding
  • SPIN –Sensor Protocol for Information via

Negotiation

“Adaptive Protocols for Information Dissemination in Wireless Sensor Networks,” Wendi Rabiner Heinzelman, J. Kulik, and H. Balakrishnan, MobiCom 1999.

10/07/2003 Ibrahim Matta

How did we review papers?

  • Motivation of the work
  • Single major idea in paper
  • Model provided in paper
  • Related work
  • Advantages of the work
  • Improvements to the work
  • Single major result
  • Future research
slide-6
SLIDE 6

10/21/2003

10/07/2003 Ibrahim Matta

SPIN

  • Motivation of the work

– Overcome limitations of classic flooding

  • Single major idea in paper

– Describe data at a high level (meta-data) and use it for negotiation – Do in-network processing to eliminate redundancy

  • Model provided in paper

– Dissemination to all sensors – meta-data smaller than data

  • Related work

– NNTP: news servers use names and timestamps as meta-data

10/07/2003 Ibrahim Matta

SPIN

Energy Dissipation:

Which one, flooding or SPIN, you expect to converge faster?

slide-7
SLIDE 7

10/21/2003

10/07/2003 Ibrahim Matta

SPIN

Broadcast Broadcast Broadcast Unicast Unicast Unicast Multicast Multicast Multicast Static

Interest Propagation

Target Detection Continuous Query Query Query Broadcast Multicast Multicast Multicast Unicast Unicast Unicast

Data Dissemination

Mobile Mobile Stationary Stationary

Event Users (Sinks) Sensors (Sources)

10/07/2003 Ibrahim Matta

SPIN

  • Advantages of the work

– Simple: ADV –REQ –DATA – scalable: only local interactions – Low latency and energy-efficient – Robust to failures and mobility

  • Improvements to the work

– Consider network losses and queuing delays

  • Single major result

– More energy efficient than flooding and close to ideal dissemination

  • Future research

– Can we do efficient dissemination without requiring all nodes to be up all the time?

slide-8
SLIDE 8

10/21/2003

10/07/2003 Ibrahim Matta

Gradient Based Approaches

  • Directed Diffusion

“Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks,” Chalermek Intanagonwiwat, Ramesh Govindan and Deborah Estrin, MobiCOM 2000.

  • GRAB –GRadient Broadcast

“GRAdient Broadcast: A Robust Data Delivery Protocol for Large Scale Sensor Networks,” Fan Ye, Gary Zhong, Songwu Lu, Lixia Zhang, ACM Wireless Networks.

10/07/2003 Ibrahim Matta

Directed Diffusion

  • Motivation of the work

– Distributed sensing and not everyone may be interested in the sensed data

  • Single major idea in paper

– Query-initiated: interests set gradients toward sink – Sink reinforces a primary (best) path – Interests refreshed periodically and aggregated inside the network

  • Model provided in paper

– Events described by attribute-value pairs – Users express interest in certain events – Probably works well for long-lived queries

  • Related work

– IP multicast: members join sessions of interest – Reliable multicast: local recovery at routers

slide-9
SLIDE 9

10/21/2003

10/07/2003 Ibrahim Matta

Directed Diffusion

Multiple Sources Link Failure Multiple Sinks Do you see any problem here?

10/07/2003 Ibrahim Matta

Directed Diffusion

Energy Dissipation:

Why is Diffusion more efficient than Omniscient Multicast?

slide-10
SLIDE 10

10/21/2003

10/07/2003 Ibrahim Matta

Directed Diffusion

Latency:

Why does Diffusion have delay comparable to Omniscient Multicast?

10/07/2003 Ibrahim Matta

Directed Diffusion

  • Advantages of the work

– Robust: only local interactions – Low latency: data received along best path – Robust: interests refreshed

  • Improvements to the work

– Diffuse interests geographically instead of flooding – Consider congestion – Data aggregation beyond suppressing duplicates – Reinforce multiple paths to avoid energy depletion on primary path

  • Single major result

– More energy efficient than flooding and omniscient multicast (source-rooted tree to all sinks)

  • Future research

– How can we reduce waste in energy due to sink-initiated reinforced paths? – Can we analyze stability of selecting reinforced paths?

slide-11
SLIDE 11

10/21/2003

10/07/2003 Ibrahim Matta

Directed Diffusion

Broadcast Unicast Multicast Static Static Static

Interest Propagation

Target Detection Continuous Broadcast Broadcast Broadcast Multicast Unicast

Data Dissemination

Mobile Mobile Mobile Mobile Mobile Mobile Stationary Stationary

Event Users (Sinks) Sensors (Sources)

Query

GRADient Broadcast: A Robust Data Delivery Protocol For Large Scale Sensor Networks

Fan Ye, Gary Zhong, Songwu Lu,Lixia Zhang UCLA ACM WINET

Niky Riga Sensor Networks Seminar Fall 2003 Boston University

slide-12
SLIDE 12

10/21/2003

10/07/2003 Ibrahim Matta

Setup…

sink

stimuli source

10/07/2003 Ibrahim Matta

Flow of data (1)

Broadcast

High energy consumption

slide-13
SLIDE 13

10/21/2003

10/07/2003 Ibrahim Matta

Flow of data (1)

Single Path

10/07/2003 Ibrahim Matta

Flow of data (3)

Multipath

Idea : Maintain more than one path from the source to the sink.

slide-14
SLIDE 14

10/21/2003

10/07/2003 Ibrahim Matta

Multipath

Disjoint Paths

For each two paths all the nodes along the path are different except the source and the sink

10/07/2003 Ibrahim Matta

Multipath

Braided Paths

Two different paths from the source to the sink differ in at least two nodes

slide-15
SLIDE 15

10/21/2003

10/07/2003 Ibrahim Matta

Braided VS Disjoint multipaths

10/07/2003 Ibrahim Matta

Proof

f : probability of node failure on a link ) (N P

d fail

: probability that a packet fail to reach the destination after N hops in disjoint

) (N P

g fail

: probability that a packet fail to reach the destination after N hops in mesh

slide-16
SLIDE 16

10/21/2003

10/07/2003 Ibrahim Matta

Proof (e=0)

N m g fail m N d fail

f N P f N P ) 1 ( 1 ( ) ( ) ) 1 ( 1 ( ) ( − − = − − =

Probabilit y t hat a packet fails t o reach hop N for different node failure rate Probabilit y t hat a packet fails t o reach hop N for different num ber of nodes m 10/07/2003 Ibrahim Matta

GRAB model

sink

stimuli source

slide-17
SLIDE 17

10/21/2003

10/07/2003 Ibrahim Matta

GRAB

  • To build paths between the source and the

sink, the sinks creates a cost field.

  • The cost at each node is the minimum

energy overhead to forward a packet from the node to the sink.

10/07/2003 Ibrahim Matta

Cost Field

  • Sink broadcasts an ADV with

cost 0

  • I nitial cost for the nodes 8
  • Each node when receives an

ADV add the cost advertised with the cost from the sender to the node. I t keeps the sm aller cost between the new and the old.

  • Event driven updates
slide-18
SLIDE 18

10/21/2003

10/07/2003 Ibrahim Matta

GRAB forwarding algorithm

Packet carries except from the data:

  • a : credit defined by source
  • Csource :cost from source to sink
  • Pconsumed:amount consumed until this node
  • Csender :the cost from current sender to sink

Source creates the packet . Each node with Creceiver<Csource calculates :

2

,         = − + = − =

source receiver thresh source receiver consumed used used

C C R C C P R α α α α

α

If R

a > R t hr esh forwards the message. 10/07/2003 Ibrahim Matta

Complexities of GRAB

  • Kept state O(n), n number of sinks
  • Packet analysis in each node O(1)
  • Energy overhead

– Forwarding O(k),k number of data reports – Cost field building proportional to #updates and #sinks

slide-19
SLIDE 19

10/21/2003

10/07/2003 Ibrahim Matta

Evaluation

Criteria

sink in received packets # source by sent packets # ratio Success =

  • Success ratio
  • Total energy consumption
  • Control packet overhead

10/07/2003 Ibrahim Matta

Evaluation

Credit a

slide-20
SLIDE 20

10/21/2003

10/07/2003 Ibrahim Matta

Evaluation

Threshold functions

4 4 3 3 2 2 1

, ,         =         =         = =

source A source A source A source A

C C f C C f C C f C C f

10/07/2003 Ibrahim Matta

Evaluation

Environmental Settings

slide-21
SLIDE 21

10/21/2003

10/07/2003 Ibrahim Matta

Evaluation

GRAB vs Directed Diffusion

10/07/2003 Ibrahim Matta

Related Work

  • Directed diffusion, braided-diffusion,

– Scalar vs. vector routing states – Receiver-decided vs. Sender-appointed – Multiple interleaving paths vs. single path

  • gradient routing

– Both receiver-decided, scalar routing states – Controllable mesh vs. cost field only

  • energy aware routing(piconet)

– Both scalar routing states – Receiver-decided vs. sender-probabilistically-appointed

slide-22
SLIDE 22

10/21/2003

10/07/2003 Ibrahim Matta

Summary

  • A forwarding mesh of multiple interleaving paths

ensures robustness

– A sender does not bind the forwarding to any specific neighbor – Multiple receivers decide on their own whether to forward

  • Per-packet credit builds mesh on-the-fly

– Dividing credit among all hops for robustness – No state maintained about which node is in mesh

10/07/2003 Ibrahim Matta

Future Work

  • Adaptive credit assignment
  • Sink inform source about data quality
  • Sink mobility
slide-23
SLIDE 23

10/21/2003

Questions??

Thank you ☺